The Montevideo machine-learning shop that has been turning data into decisions since before the world learned to say "AI" out loud.
Walk into a video call with Tryolabs in 2026 and you will not hear the word "revolutionary." You will hear questions. What is in your data? What decision are you actually trying to make? What happens to this model on a Tuesday at 3 a.m. when nobody is watching? The company built its name on a phrase most of its industry avoids - no hype - and somehow that restraint became its loudest selling point.
For sixteen years this firm has done the unglamorous middle of artificial intelligence: the part between a flashy demo and a system that actually earns its keep. It ships object detectors that count fish at sea, pricing engines that move retail margins, and predictive models that flag a failing oil well before it becomes a headline. The demos rarely go viral. The invoices get paid.
Tryolabs sits on the Rambla in Montevideo, an ocean and a hemisphere away from the Bay Area boardrooms it has quietly served since the scrappy early days of machine learning. It was, as the company likes to remind people, doing this before "AI" was a line item in anyone's budget.
It names its open-source tools after the Metroid video-game universe. Norfair. Luminoth. A small joke, shipped to thousands of GitHub stars.
Tryolabs is a consultancy, not a product company - which means the "product" is whatever your problem requires. Here is the work, in plain terms.
Object detection, multi-object tracking and image recognition - from retail shelves to fishing boats.
LLM-powered copilots, workflow automation and agents built to survive contact with production.
Dynamic pricing, demand forecasting and inventory optimization for e-commerce and retail.
Scaling, monitoring and lifecycle management so a model keeps working long after launch day.
Pipelines, data lakes and migrations - the unsexy plumbing that every model secretly depends on.
Roadmaps, change management and AI Centers of Excellence for teams figuring out where to even start.
A Uruguayan team of fewer than a hundred people building the AI behind some very large logos - across retail, airlines, energy, insurance and the nonprofit world.
Reputation in AI is hard to fake when your code is public. Tryolabs' open-source libraries are its quiet resume - downloaded, forked and starred by engineers who will never sign a contract but will absolutely judge the work.
Norfair adds real-time multi-object tracking to any detector in a few lines of Python. Luminoth was an early deep-learning toolkit for computer vision. Requestium stitches together Requests and Selenium for web automation. Note the naming: Norfair and Luminoth are both lifted from the Metroid universe.
Approximate star counts; figures move over time. Luminoth is no longer actively maintained - the team now points users toward more modern detectors.
Somewhere on the open ocean, a camera on a fishing vessel is doing math. It is counting catch, identifying species, and doing it without a data center anywhere in sight. That work came out of a Tryolabs collaboration with The Nature Conservancy - edge AI applied to fisheries monitoring, where connectivity is poor and the stakes for marine ecosystems are not.
The firm files this under "AI for Good," alongside engagements touching global development and the kind of social-impact problems that rarely make pitch decks. It is a useful tell about the culture: the same object-tracking instincts that optimize a retail aisle get pointed at a coral reef.
Three founders - Raul Garreta, Martin Alcala Rubi and Ernesto Rodriguez Di Paolo - start Tryolabs in Montevideo, building Python/Django apps with machine learning for ambitious startups.
Launches Luminoth, an open-source computer-vision toolkit built on TensorFlow.
Marks 10 years - a decade of applied AI from Latin America to the world.
Releases Norfair, an open-source library for real-time object tracking.
Ships Norfair 2.0 - the biggest upgrade to its tracking library since launch.
Company materials cite a leader recognized among TIME100 AI influencers and a Google Cloud Gemini Enterprise practice.
Machine-learning entrepreneur; also co-founded the text-analysis company MonkeyLearn.
Recognized by the Uruguayan Engineering Association for advancing AI software exports from Uruguay.
Part of the founding trio that started Tryolabs in 2009.
Open-source projects are named after locations from the Metroid universe. Norfair is a literal place in the game.
It was building ML products in 2009 - before "AI" was a boardroom buzzword anyone wanted to claim.
Headquartered on Montevideo's Rambla, it serves clients in San Francisco and beyond.
Its public motto leans on "no hype" - rare honesty in the generative-AI era.
Search links point to live results; specific videos may vary over time.
Return to that call. The one with no fireworks and a lot of questions. By the end of it, something has shifted - not in the slide deck, but in the room. The retailer leaves with a pricing model instead of a vision statement. The energy company leaves with an early-warning system instead of a slogan. The conservation group leaves with a camera that can see fish.
That is the trick Tryolabs has been quietly running since 2009: it changed the conversation about AI by refusing to oversell it. While the rest of the industry mastered the demo, this Montevideo shop mastered the part that comes after - the Tuesday at 3 a.m., the invoice that gets paid, the model that still works. No hype. Just the work, shipped.